Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74239.71051296088,
-      "required_pages_per_second_upper_estimate": 0.006434909948437386
+      "required_token_positions_per_second": 75403.66849622988,
+      "required_pages_per_second_upper_estimate": 0.006535798876402604
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9279.96381412011,
-      "required_pages_per_second_upper_estimate": 0.0037019163132759334
+      "required_token_positions_per_second": 9425.458562028734,
+      "required_pages_per_second_upper_estimate": 0.003759956343557019
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18559.92762824022,
-      "required_pages_per_second_upper_estimate": 0.004293512613223381
+      "required_token_positions_per_second": 18850.91712405747,
+      "required_pages_per_second_upper_estimate": 0.004360827911840487
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27839.891442360327,
-      "required_pages_per_second_upper_estimate": 0.0016953033242178147
+      "required_token_positions_per_second": 28276.375686086205,
+      "required_pages_per_second_upper_estimate": 0.0017218829246048787
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27839.891442360327,
-      "required_pages_per_second_upper_estimate": 0.002143191908984838
+      "required_token_positions_per_second": 28276.375686086205,
+      "required_pages_per_second_upper_estimate": 0.0021767936743325752
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27839.891442360327,
-      "required_pages_per_second_upper_estimate": 0.34567827759117337
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28276.375686086205,
+      "required_pages_per_second_upper_estimate": 0.3510979510794567
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11302162930473258,
-    "producer_p95_seconds": 0.2140635759997167,
-    "gpu_update_seconds": 0.9931073207394641,
-    "producer_batches_per_second": 8.8478639544628,
-    "gpu_batches_per_second": 1.0069405180251856,
-    "producer_headroom": 8.78687846608383,
+    "producer_batch_seconds": 0.08872417996874304,
+    "producer_p95_seconds": 0.20614324500002112,
+    "gpu_update_seconds": 0.9777773611065931,
+    "producer_batches_per_second": 11.270884671487451,
+    "gpu_batches_per_second": 1.0227277085534652,
+    "producer_headroom": 11.02041587142375,
     "sustainable": true,
-    "buffer_seconds": 60.57954656510731,
+    "buffer_seconds": 60.62219638860877,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362290.61797311145,
+    "generation_bytes_per_second": 362230.3766387865,
     "upload_bytes_per_second": 5902263.000178364,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.73913292307861,
+    "estimated_full_shards_per_hour": 77.72620653507897,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 122500, "loss": 0.5423185117542744, "lr": 0.0003, "grad_norm": 0.18062500836188222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17926705528, "tokens_per_second": 149139.726335044, "data_wait_s": 1.0955999641737435e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 122600, "loss": 0.6145331133157015, "lr": 0.0003, "grad_norm": 0.1591392239095421, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17941202218, "tokens_per_second": 147406.74475259118, "data_wait_s": 1.597299979039235e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 122700, "loss": 0.6094257552176714, "lr": 0.0003, "grad_norm": 0.13699610787656766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17955682394, "tokens_per_second": 148236.24361761165, "data_wait_s": 1.1531999916769564e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 122800, "loss": 0.5962003283202648, "lr": 0.0003, "grad_norm": 0.1375012183929598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17970401412, "tokens_per_second": 149914.04022148447, "data_wait_s": 1.2873999367002398e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33c8f5b30578159d6a83ef714012e2e854c7c8249a957a0a62e1452502a11d2f
-size 164034
+oid sha256:401bb2f7ee2a4443159f47501d7870ee328f268f4684e0cd194fe0b257d41617
+size 166152